BETWEEN INNOVATION AND SAFETY: THE ROLE OF BILL 2338 (2023) IN REGULATING ARTIFICIAL INTELLIGENCE IN BRAZIL

Regulating a technology in its early stages is always a complex challenge. At this phase, data on risks and social impacts are still scarce, making it difficult to draft precise norms. This is the well-known Collingridge dilemma: initially, there is little information but ample regulatory space; after diffusion, there is abundant information but limited scope for regulatory interventions.

Regarding Artificial Intelligence (AI), we face a global challenge that places safety and technological development in conflict. Despite efforts to reconcile both sides, differing worldviews are likely to persist, making consensus among "experts" on AI risks improbable.

In Brazil, we are monitoring the progress of Bill 2338 (2023), which originated in the Federal Senate following extensive discussions led by the CJSUBIA—Commission of Jurists responsible for supporting the drafting of a substitute bill on artificial intelligence in the country.

In a recent article on the subject, Arvind Narayanan and Sayash Kapoor presented their view of AI as a "normal" technology, distancing themselves from utopian and dystopian perspectives—without underestimating its impact. But what does this mean in practice? The pace of innovation differs from the pace of adoption and diffusion of a technology. The latter is limited by the capacity of individuals, organizations, and institutions to adapt to the technology.

Practically speaking, understanding AI as a normal technology (and not as a superintelligence that will dominate the world) allows regulatory work to be conducted more cautiously, as the technology will progressively integrate into social and economic life—similar to other general-purpose technologies like electricity. This opens the door to a resilient regulatory approach that combines anticipatory planning with continuous adaptation.

Gary E. Marchant and Yvonne A. Stevens discussed this in “Resilience: A New Tool in the Risk Governance Toolbox for Emerging Technologies” predicting that traditional ex-ante risk management approaches face difficulties in providing adequate governance for emerging technologies due to the challenge of forecasting realistic risk scenarios.

Examining Bill 2338 (2023) reveals a predominance of ex-ante governance measures, such as: risk classification (Art. 12); mandatory documentation (Art. 18, I, items “a” through “d”; Art. 18, II, items “a” through “c”); measures to mitigate and prevent discriminatory biases (Art. 18, I, item “e”); transparency requirements (Art. 18, II, item “f”); auditability (Art. 18, II, item “b”); interpretability (Art. 18, II, item “d”); and algorithmic impact assessment (Art. 25).

Many of these ex-ante obligations parallel the AI Act, demonstrating the strong inspiration the Brazilian proposal draws from the European document. An exploratory study by the Institute for Technology and Society of Rio found that, among the 39 obligations described in the European framework, 13 have counterparts in the Brazilian framework, indicating a comparability of about one-third of the European obligations with equivalents in the national proposal.

However, it is necessary to acknowledge the profound differences between the European and Brazilian contexts in AI development. Brazil announced the Brazilian Artificial Intelligence Plan (PBIA) 2024–2028, projecting an investment of $4 billion over four years. Meanwhile, the European Union launched InvestAI, with an expected investment of $216 billion in the coming years. In contrast, the United States accumulated approximately $474 billion in investments between 2013 and 2024, while China invested around $119 billion in the same period.

Source: Artificial Intelligence Index Report 2025. Stanford Institute for Human-Centered Artificial Intelligence, 2025.

As mentioned at the outset, understanding AI as a technology that will progressively integrate into social and economic life does not mean underestimating its risks. This article does not advocate for the complete absence of ex-ante governance obligations, favoring development per se regardless of safety.

Nevertheless, Brazil's regulatory decision must consider the difficulties the country will face in the global competitive landscape—especially when Bill 2338 (2023) commendably provides for ex-post measures that protect human rights while allowing technological development.

Mechanisms such as continuous monitoring of AI systems (Art. 15, § 1), reporting of serious incidents (Art. 42), supervision and oversight (Arts. 45 to 52), and the experimental regulatory environment (Art. 55), also known as a regulatory sandbox, promote technological advancement while enabling authorities to act in real risk scenarios (not just speculative ones), mitigating unforeseen damages and adverse outcomes.

It is also important to note that Bill 2338 (2023) introduces a significant set of rights for individuals and groups affected by high-risk AI systems (Art. 6), including the right to explanation, contestation, and human review — in particular, human oversight (Art. 8) — aiming to prevent or minimize risks to human rights.

These ex-post governance measures allow for technological development without compromising human rights, positioning Brazil in the global regulatory landscape, especially considering the country's already lower volume of investments in the area.

Hindering the development of national solutions, which could reduce our dependence on major technological powers, through regulatory impact, does not seem to be a desirable path for a developing country seeking prominence on the world stage.

The regulation of artificial intelligence in Brazil, as proposed in Bill 2338 (2023), represents a significant advancement in building a normative framework aimed at protecting human rights and promoting technological innovation. However, for this effort to yield the desired effects, it is essential that the country adopts a balanced regulatory approach that considers both the specificities of the national context and the global asymmetries in technology development and financing.

Sobre o autor: Ricardo de Souza Mello Filho

Pesquisador do LABSUL

Mestre em Direito pela UNESC

Sócio Baldin Mello

Advogado, Professor e Consulto